Uber blew through its entire 2026 AI budget by mid-April. Four months into the year, the well was dry, and the company’s top executives are now publicly asking a question that a lot of tech leaders are thinking privately: is any of this actually working?

Uber President and COO Andrew Macdonald put it bluntly in a recent interview. The company can’t draw a clear line between skyrocketing AI token consumption and the delivery of meaningfully better features to consumers. Adoption metrics look incredible on paper, but the real-world output? Fuzzy at best.

The numbers tell a strange story

Here’s what happened. Uber rolled out Anthropic’s Claude Code to roughly 5,000 engineers back in December 2025. Adoption was volcanic. Usage of agentic coding features surged from 32% in February to 84% by March 2026. By that point, 95% of Uber’s engineers were using AI tools on a monthly basis, and nearly 70% of code commits involved them in some capacity.

Macdonald said in a May 2026 interview that the connection between rising token consumption and useful consumer features simply isn’t there yet. “It’s very hard to draw a line between one of those stats and, ‘Okay, now we’re actually producing 25 percent more useful consumer features,'” he noted.

Meanwhile, per-engineer monthly API costs ballooned to between $500 and $2,000, blowing past internal forecasts. CTO Praveen Neppalli Naga confirmed that the full annual AI budget was exhausted by mid-April. For context, Uber’s total R&D spending hit $3.4 billion in 2025, a 9% increase year-over-year.

What investors should watch

Third, consider the pricing dynamics. Per-engineer API costs of $500 to $2,000 per month are eye-watering, and they suggest that AI inference providers, including Anthropic, hold substantial pricing power in the current market. That’s great for compute providers in the short term, but unsustainable costs tend to trigger one of two responses: companies either find cheaper alternatives, which is where decentralized compute could theoretically benefit, or they simply use less.

The broader lesson from Uber’s AI budget blowout is one the crypto industry should internalize. Adoption is not the same as value creation. Usage metrics are not the same as productivity gains. And spending money faster than planned is not, in itself, evidence that something is working. Uber’s candid admission that the link between spending and results “is not there yet” should land as a meaningful signal, not just background noise.